Dedicated License Plate Cam project

[nayr]

this has been on my todo list for a good while, I think my house is in a pretty good location as there is a bend 2 houses up hill that points people's plates directly at my roof line.. but I might need better than the 25mm, leme us know how far away you can reach with that setup, sadly as we lost Carl before he completed his report..

 

[blake]

I'm thinking I need to add this to the todo list also. Since there are soooo many HOA's around here that I could offer it to!

 

[pozzello]

this is a 2032 (originally 12mm) hik bullet with the wrightwood 25mm lens, from about 60 feet:

I've got a 15 degree IR illuminator on the way to help with the night shots...

 

[nayr]

yeah im going to need a much bigger zoom lol, just playing around with the measuring tool on google maps and I remember why I shelved this project.. I have about 160ft to cover before I hit that nice curve, hell even the intersection dead infront of my house going to a culdesac is 75 ft away from my structure.. but traffic to the culdesac is minimal no point in ALPR that road.

 

[Git]

A couple of links to BlueIris Exports I made this morning

Hik 2032 with 12mm lens

Huisun Mini PZT v1 with 10x zoom (FedEx just dropped something off and was turning around)


The Huisun is obviously zoomed in closer - but I found out that a lot of people like to cut this corner and drive on the wrong side of the road (short dead end street). If they stayed to the left of the man hole cover (where they should be coming towards me) then I would not be able to get their plate with the Huisun where I could with the Hik. It's a trade off since a lot of cars do not have front license plates, so now I focus mainly on getting a rear plate and if I do get the front it is a bonus. (In the Hik video - I just temporarily set it up this morning and it could have been aimed to the left quite a bit more to get the whole street)

A couple of more

UPS Truck - Hik 4mm

UPS Truck - Husisun 10x @ zoomed into 6x

 

[ruppmeister]

Here is a link showing what a 12 mm 2032 looks like in daylight at about 70'

Haha, how did you get The Stig to drive that Honda in your demo video? Plates looks legible to me

 

[steve457]

I'm debating whether I need to get one or two cameras to cover the street. I'm going to try it with one camera first positioned like the following diagram.
I'm not sure whether covering both streets will allow me to see plates clearly or not, so will need to try it out. Another part of the challenge is the small tree (green dot), which will partially block the camera view. If I were to use two cameras I could avoid the tree altogether by aiming one down each street, but would need a bit more zoom. I'll post back how it goes once my camera arrives.

 

[Git]

I edited my post and added a couple of more examples

I am currently running 2 cameras - but the 4mm Hik is more of an "overview" and to cover our car in the driveway and is also set to trigger the other camera when it detects motion

 

[nayr]

Here's some shots of what I've got to work with.. lets not turn this into a political discussion pls

Intersection to a dead end, but that would be easy peasy.. the pan/tilt camera that took these shots is being replaced by one with 4x optical zoom (12mm) with a better horizon so I am hoping to find a bit of success with that..

I want to get em as they go around that bend, and I'll have to put a camera higher up and further right to get that tree trunk out of view and not have issues with people parked on side of street.. with google maps I'm looking at 150ft likely...

I am thinking my distances might be a bit too far for the cheap 25mm hack.. suspect i need to find a 50-60mm lens, any suggestions? this project needs to be really cheap because its actual value to me wont be that high, it'll help the neighborhood watch more than me I suspect

 

[klasipca]

A couple of links to BlueIris Exports I made this morning

Hik 2032 with 12mm lens

Huisun Mini PZT v1 with 10x zoom (FedEx just dropped something off and was turning around)


The Huisun is obviously zoomed in closer - but I found out that a lot of people like to cut this corner and drive on the wrong side of the road (short dead end street). If they stayed to the left of the man hole cover (where they should be coming towards me) then I would not be able to get their plate with the Huisun where I could with the Hik. It's a trade off since a lot of cars do not have front license plates, so now I focus mainly on getting a rear plate and if I do get the front it is a bonus. (In the Hik video - I just temporarily set it up this morning and it could have been aimed to the left quite a bit more to get the whole street)

A couple of more

UPS Truck - Hik 4mm

UPS Truck - Husisun 10x @ zoomed into 6x

What about at night, does Hik get the license plate?

 

[Git]

As I recall - the Hik 12mm did a decent job, but as you can see it was not zoomed in as far as the Huisun. I remember having to play around with the videos - slowing them down, going forward and backward to try to make out all the letters. I don't usually have to do that with the Huisun For me, the main advantage of the Huizun - was the ability to play around with zoom to find the 'sweet' spot that worked for me. Obviously the more you can zoom in and still capture what you need at this price point is an advantage

I took a chance - the Hik is proven and worked versus the newcomer - the Huisun. At this point the Huisun has worked out very well, but time will tell

 

[nayr]

I think that Huisun might do the trick, thx for the vids..

*edit*
Upon further research, while the HuniSun looks like a great solution for ALPR it is lacking critical API required for an easy implementation..

problems: getting jpeg snapshots are not straight forward, and they purposely locked down API, so getting motion events will be hard.

Had it done like Dahua or Hikvision with a standard and open HTTP API, would pretty easy for me to detect a car going by, grab the jpeg, pass it onto ALPR and store the results in my automation database... but no, they had to be stupid.. now i have to process a rstp stream for motion and difficulty to implement just got a bit harder and required resources went up.

Does this implement ONVIF well? that might be an option but it'd have to be specific.. What are the network storage options on motion? can it dump to a ftp, guess I might be able to catch the file, convert to a bunch of jpegs and then send it through alpr.

 

[Kawboy12R]

I know this doesn't help you with a solution that integrates into your current system but the ANPR software mentioned earlier works with my Huisun miniptz.

 

[nayr]

Can it save a MJPEG to FTP on motion without missing the car? I am thinking KISS, I could watch a folder for incoming mjpegs, very quickly convert it to a series of jpg files, run them all trough the ALPR engine.. pick the image with the highest results over a threshold and store it in database with raw data.. finally cleanup whats left behind and then wait for the next video file to appear in the folder.

If wrote an application to do this it would be pretty much universal I think, would not rely on any API calls and the storage problems with MJPEG would be irrelevant since I would just be using it for temporary storage thats easier to decode.

This camera would be dedicated to nothing but ALPR and saving only the best still image it thinks it found, so at night I would be fine with a underexposed image.. might even run it at the lowest resolution I can so processing requirements are minimal.. all I really want is text/timestamp and a general description from the photo.

 

[Kawboy12R]

Don't know. Haven't played with it much beyond getting it to recognize plates. I'm not at home now either so can't play with it.

 

[GBot]

I've got a 15 degree IR illuminator on the way to help with the night shots...

Looking forward to seeing your results since my 25mm lens just got delivered and I suspect I'll need supplemental IR to pick up plates at night. Rebranded 2032 (Swann) here too, BTW.

Sent from my Nexus 5 using Tapatalk

 

[nayr]

well I installed ALPR on a Debian VM and started playing with it.. I am very happy to report the project has progressed alot in the last few years since I seriously looked at it.. its ready to go.. compiled it from source, configured the daemon to take a MJPEG D1 sub-stream from one of my dahuas and as soon as I had an old 1975 Colorado license plate in focus I got a bunch of snapshots and logs..

root@alpr:/var/log# tail -f alprd.log 
INFO - Running OpenALPR daemon in daemon mode.
INFO - Using: /etc/openalpr/alprd.conf for daemon configuration
INFO - Using: /var/lib/openalpr/plateimages/ for storing valid plate images
INFO - country: us -- config file: /etc/openalpr/openalpr.conf
INFO - Stream 1: http://south-ipc/axis-cgi/mjpg/video.cgi?camera=1
INFO - Starting camera 1
INFO - Video stream connecting...
INFO - Video stream connected
DEBUG - Writing plate AG86Z6 (test-cam1-1456113787639) to queue.
DEBUG - put job id: 2
DEBUG - Writing plate AG8626 (test-cam1-1456113791890) to queue.

root@alpr:/var/lib/openalpr/plateimages# ls
test-cam1-1456113787639.jpg  test-cam1-1456113796212.jpg  test-cam1-1456113800430.jpg  test-cam1-1456113806169.jpg  test-cam1-1456113810610.jpg  test-cam1-1456113814694.jpg
test-cam1-1456113791890.jpg  test-cam1-1456113797034.jpg  test-cam1-1456113802064.jpg  test-cam1-1456113807004.jpg  test-cam1-1456113811446.jpg  test-cam1-1456113815531.jpg
test-cam1-1456113793650.jpg  test-cam1-1456113797927.jpg  test-cam1-1456113804485.jpg  test-cam1-1456113807845.jpg  test-cam1-1456113813038.jpg  test-cam1-1456113816349.jpg
test-cam1-1456113795382.jpg  test-cam1-1456113798818.jpg  test-cam1-1456113805330.jpg  test-cam1-1456113809637.jpg  test-cam1-1456113813864.jpg  test-cam1-1456113817123.jpg

Got it on the second try so I guess I'll get a HuniSun to accomplish that 50mm zoom I am looking for.. I dont really care about its API anymore or lack of anything but windows support right now.. just want a simple MJPEG stream at low resolution.. I'll use this same VM and it'll be golden.

Wont take much to get it to grab the text data and put it into my automation server logs, just include the filename so I can find the image if nessicary and setup a script to clean out old jpegs after a few months.. I didnt know it would be this easy.

attached is the actual image it saved of the matching plate, and it kept on saving it until I took it away heh.

 

[steve457]

Great results! I'm going to start messing around with openalpr on my Ubuntu instance.

What resolution is your sub-stream on your camera?

 

[pozzello]

i guess at D1 rez and a decent processor it may be able to keep up with 10+ fps, but i'll let you figure that out.
i think that's been it's achilles' heel...

 

[pozzello]

he said D1, which is 720x480 pixels (NTSC) or 720x576 pixels (PAL).